FIGURE 4.

Construction of a prognostic risk model based on metabolism-related genes (A) The differential enrichment score of gene sets was calculated between each of the two subgroups and intersected them. Subtype 1, subtype 2 and subtype 3 had 1,098, 662 and 647 distinct gene sets, respectively. (B) Batch Cox regression analyses screening for prognosis-related differential genes. (C) The dotted vertical lines represent the optimal values of λ. The top x-axis has the numbers of gene sets, whereas the lower x-axis revealed the log (λ). (D) Least absolute shrinkage and selection operator (LASSO) coefficient profiles (y-axis) of the gene sets and the optimal penalization coefficient (λ) via 10-fold cross-validation based on partial likelihood deviance. (E) Take the intersecting genes after lasso and random-forest screening. (F) Constructing a stepwise Cox proportional hazards model. (G) Kaplan–Meier OS curves with difference detection by log-rank test for patients from the training datasets. (H) ROC curve analyses based on the 11 gene signature. (I) Relative expression levels of mRNA for 11 gene signature. (J) Coefficient of variation of 11 gene signature in cancer and paracancerous tissues.